On Sun, Jun 8, 2025 at 12:13 PM Anuj gupta <anuj1072538@xxxxxxxxx> wrote:
>
> On Sat, Jun 7, 2025 at 5:12 AM Joanne Koong <joannelkoong@xxxxxxxxx> wrote:
> >
> > This series adds fuse iomap support for buffered writes and dirty folio
> > writeback. This is needed so that granular dirty tracking can be used in
> > fuse when large folios are enabled so that if only a few bytes in a large
> > folio are dirty, only a smaller portion is written out instead of the entire
> > folio.
> >
> > In order to do so, a new iomap type, IOMAP_IN_MEM, is added that is more
> > generic and does not depend on the block layer. The parts of iomap buffer io
> > that depend on bios and CONFIG_BLOCK is moved to a separate file,
> > buffered-io-bio.c, in order to allow filesystems that do not have CONFIG_BLOCK
> > set to use IOMAP_IN_MEM buffered io.
> >
> > This series was run through fstests with large folios enabled and through
> > some quick sanity checks on passthrough_hp with a) writing 1 GB in 1 MB chunks
> > and then going back and dirtying a few bytes in each chunk and b) writing 50 MB
> > in 1 MB chunks and going through dirtying the entire chunk for several runs.
> > a) showed about a 40% speedup increase with iomap support added and b) showed
> > roughly the same performance.
> >
> > This patchset does not enable large folios yet. That will be sent out in a
> > separate future patchset.
> >
> >
> > Thanks,
> > Joanne
>
> Hi Joanne,
>
> I tried experimenting with your patch series to evaluate its impact. To
> measure the improvement, I enabled large folios for FUSE. In my setup,
> I observed a ~43% reduction in writeback time.
>
> Here’s the script[1] I used to benchmark FUSE writeback performance
> based on the details you shared: It formats and mounts an XFS volume,
> runs the passthrough_hp FUSE daemon, writes 1MB chunks to populate the
> file, and then issues 4-byte overwrites to test fine-grained writeback
> behavior.
>
> If I’ve missed anything or there’s a better way to evaluate this, I’d
> really appreciate your input — I’m still getting up to speed with FUSE
> internals.
Hi Anuj,
Thanks for testing it out locally and sharing the benchmarks. Your
test is pretty much the same as mine (except the underlying filesystem
I used for passthrough is ext4). I saw roughly a 40% speedup as well
for buffered writes.
My main concern was whether iomap overhead ends up slowing down writes
that don't need granular dirty tracking. I didn't see any noticeable
difference in performance though when I tested it out by writing out
all entire chunks.
>
> [1]
>
> #!/bin/bash
> set -e
>
> DEVICE="/dev/nvme0n1"
> BACKING_MNT="/mnt"
> FUSE_MNT="/tmp/fusefs"
> CHUNK_MB=1
> TOTAL_MB=1024
> DIRTY_BYTES=4
> REPEATS=5
> LOGFILE="fuse_test_results.csv"
> DIR=$(date +"%H-%M-%S-%d-%m-%y")
>
> mkdir $DIR
> echo "$DIR created"
>
> mkdir -p "$BACKING_MNT" "$FUSE_MNT"
>
> echo "run,duration_seconds" > "$LOGFILE"
>
> for run in $(seq 1 $REPEATS); do
> echo "[Run $run] Formatting $DEVICE with XFS..."
> mkfs.xfs -f "$DEVICE"
>
> echo "[Run $run] Mounting XFS to $BACKING_MNT..."
> mount "$DEVICE" "$BACKING_MNT"
>
> echo "[Run $run] Starting passthrough_hp on $FUSE_MNT..."
> ./passthrough_hp --nopassthrough "$BACKING_MNT" "$FUSE_MNT" &
> sleep 2
>
> echo "[Run $run] Dropping caches and syncing..."
> sync
> echo 3 > /proc/sys/vm/drop_caches
>
> TEST_FILE="$FUSE_MNT/testfile_run${run}"
>
> for ((i=0; i<$TOTAL_MB; i++)); do
> dd if=/dev/urandom bs=1M count=1 oflag=direct seek=$i
> of=$TEST_FILE status=none
> done
>
> START=$(date +%s.%N)
> for ((i=0; i<$TOTAL_MB; i++)); do
> offset=$((i * 1048576 + 1048572))
> #offset=$((i * 1048576))
> dd if=/dev/urandom bs=1 count=$DIRTY_BYTES of=$TEST_FILE
> seek=$offset status=none
> done
>
> fusermount -u "$FUSE_MNT"
> umount "$BACKING_MNT"
>
> END=$(date +%s.%N)
> DURATION=$(echo "$END - $START" | bc)
> echo "$run,$DURATION" >> $DIR/"$LOGFILE"
> echo "[Run $run] Duration: ${DURATION}s"
> done
>
> echo "All runs complete. Results saved to $DIR/$LOGFILE."
